Text reasoning problem processing optimization method and system

The invention discloses a text reasoning problem processing optimization method and system, and the method comprises the steps: obtaining a text reasoning problem data set without thinking chain labeling, carrying out the thinking chain labeling of the text reasoning problem data set without thinking chain labeling through a large language model, and obtaining a first thinking chain data set; wherein each group of data in the first thinking chain data set comprises a question, a plurality of thinking chains and a plurality of answers; calculating the entropy of each answer in each group of data, and filtering the data of which the entropy is greater than a preset filtering value to obtain a second thinking chain data set; obtaining a test problem, and retrieving a plurality of pieces of data related to the semantics of the test problem from the second thinking chain data set through a double-tower model to serve as a candidate group; and obtaining a test question reference group through a large language model